Follow the instructions for adding a new production metric, and point collection at stat1009 with the expected port and path, probably :9568/metrics .
Alternatively: it's also possible to set up ephemeral job push metrics but these have the major drawback of not easily capturing runtime performance metrics, which is the main raison d'etre of our metrics.
When done, update our internal documentation and remove the deprecated Prometheus process. This might still be useful for local dev, so it can be ported to the code readme.